Quantum chemistry and wavefunction based methods for ... · 7/12/2011 · The 1998 Nobel Prize in...
Transcript of Quantum chemistry and wavefunction based methods for ... · 7/12/2011 · The 1998 Nobel Prize in...
HUMBOLDT-UNIVERSITÄT ZU BERLIN
Quantum chemistry and wavefunction basedmethods for electron correlation
Hands-on Tutorial Workshop „Ab initio Molecular Simulations“Fritz-Haber-Institute, Berlin, July 12, 2011
Joachim SauerInstitut für Chemie, Humboldt-Universität
©Joachim Sauer, HU Berlin, 2011
The Royal Swedish Academy of Sciences has awarded The 1998 Nobel Prize in Chemistry in the area of quantum chemistry to
Walter Kohn, University of California at Santa Barbara, USA andJohn A. Pople, Northwestern Univ., Evanston, Ill., USA (British citizen).
Citation:"to Walter Kohn for his development of the density-functional theory and toJohn Pople for his development of computational methods in quantumchemistry."
©Joachim Sauer, HU Berlin, 2011
Surface Reaction
TS
Catalyst+ Educt
IntrinsicBarrier
Catalyst-EductComplex
Catalyst-ProductComplex
Bindingenergy
Reaction energy
ApparentBarrier
©Joachim Sauer, HU Berlin, 2011
DFT problems with barriers
TS
Catalyst+ Educt
IntrinsicBarrier
Catalyst-EductComplex
Catalyst-ProductComplex
Bindingenergy
Van der Waals(dispersion)
SI error(reaction site)
©Joachim Sauer, HU Berlin, 2011
Zeolite catalysts: active sites (transition metal ions, protons)in a „surface-only“ silica matrix
H
-Si+Al,
AlSiO O
O O
AlSiO O
O O
H
SiSiO O
O OM+
H
-Si+Al,
AlSiO O
O O
AlSiO O
O O
H
SiSiO O
O OM+
©Joachim Sauer, HU Berlin, 2011
Zeolite catalysis: Methanol-to-hydrocarbons
Review: Stöcker, Microp. and Mesop. Mat. 29 (1999) 3-48
Methanol fromdifferent sources
CH3OH CH3OCH3
CH2=CH2
CH2=CH-CH3↓
Cyclic Polyens
Aromatic HC
Methylation
C-C Formation, induction
↓
↓
↓
+ CH3OH
GasolineOlefines
MTGMTO
Hydrocarbon poolmechanism
Haw et al, Kolboe et al.
©Joachim Sauer, HU Berlin, 2011
MeOH (g)Alkene (g)
MeOH (ads)Alkene (g)
MeOH (ads)Alkene (ads) Water (ads)
Alkene (ads)
Water (g)Alkene (g)
Transitionstructure
*S. Svelle, P.A. Ronning, S. Kolboe, J. Catal. 2004, 224, 115.S. Svelle, P.O. Ronning, U. Olsbye, S. Kolboe,J. Catal. 2005, 234, 385.
Methylation of ethene, propene, trans-2-butene in H-MFI
Measured* Apparent BarrierEthene 104Propene 64 (-40)t-2-Butene 40 (-24)
©Joachim Sauer, HU Berlin, 2011
Methylation of alkenes in H-MFI - pbc DFT
Svelle, Tuma, Rozanska, Kerber, Sauer, JACS 131 (2009) 816
Error increases with chain length
©Joachim Sauer, HU Berlin, 2011
s6 = 1.4/1.3/0.7 for BLYP/BP86/PBE
Parameters are as good for solids (condensed systems) as formoleculesNote, however, Mg2+ very different from Mg, whereas O2- similar to O
Implementation of Ewald sum for 1/r6 for periodic systems
Etotal = EDFT +Edisp
Edisp = !s6C6
ij
rij6j = i +1
N
"i =1
N !1
" fdamp fdamp (R) = 1
1+e-! (R/R0 -1)
Pragmatic solution: DFT + Dispersion (DFT+D)
Grimme, J. Comput. Chem., 2004, 25, 1463; 2006, 27, 1787.
Many predecessors with DFT, HF+Disp, e.g. Ahlrichs and ScolesMany more sophisticated schemes
Kerber, Sierka, Sauer, J. Comput. Chem. 29 (2008) 2088.
©Joachim Sauer, HU Berlin, 2011
Methylation of alkenes in H-MFI - pbc DFT+D
Svelle, Tuma, Rozanska, Kerber, Sauer, JACS 131 (2009) 816
Dispersion - increases with chain length
©Joachim Sauer, HU Berlin, 2011
Methylation of alkenes in H-MFI - pbc DFT+D
Svelle, Tuma, Rozanska, Kerber, Sauer, JACS 131 (2009) 816
Dispersion
Too low barrierconstant SIC error
©Joachim Sauer, HU Berlin, 2011
Errors on energy barriers (Truhlar et al.)
Nucleophilic substitution
kcal/mol
Reference data:CCSD(T) with extrapolation to complete basis set limit(„W1 theory“)
©Joachim Sauer, HU Berlin, 2011
DFT (PBE/plane waves), VASP20.2x20.5x13.5 Å; 96 +18 atomsPeriodic Boundary Conditions
b _
_a
CCSD(T)/cbs17+17 AtomeC3O11Si2AlH17MOLPRO
MP2/TZVP123+67AtomeC3O72Si47AlH67CC2 code
Divide and Conquer - Models and Methodsn
©Joachim Sauer, HU Berlin, 2011
Hybrid high level : low level method
Ehybrid (S,C) = EDFT+D(S) + [EMP2(C) - EDFT+D(C)]
High level correction
Step 0: PBE+D optimization, periodic boundary conditions (pbc) Frequency calculation for stationary points, ZPVE
Step 1: Hybrid MP2(cluster):PBE+D(pbc) optimization[Step 2: Basis set extrapolation to CBS limit, single point Step 3: CCSD(T)-MP2, small cluster model
Tuma, Sauer, CPL 2004, 387, 388; PCCP, 2006, 8, 3955Kerber, Sierka, Sauer, J. Comput. Chem. 2008, 29, 2088Reuter/Scheffler, PRL 98 (2007) 176103 (CO/Cu(111))
Stoll, JPC A 113 (2009) 11483 (Be, Mg crystals)
Hybrid MP2(cluster):PBE+D(pbc) +ΔCCSD(T) method
© Joachim Sauer, HU Berlin, 2011
Energy as functional of orbitals or electron density
kinet. e-n e-e e-e en. attract Coulomb exchange/corr orbital density density
EHF = ET + EN + EJ + EXFock30 (orbital)
EDFT = ET + EN + EJ + EXC (density)functional of density onlyEXC
LDA = EXDirac30 + EC
VWN
functional also of density gradient (Generalized Gradient Approximation)EXC
BLYP = EXDirac30 + ΔEX
B88 + ECLYP
Self-interactioncancels
© Joachim Sauer, HU Berlin, 2011
12 !(x1) 1r12"" !(x2 )dx1dx2 # 1
2 i*(x1) j$(x2 )"" 1
r12j (x1)i (x2 )dx1dx2
i, j%!(x) =
i(x)i=1
N
" i*(x)
Coulomb and exchange terms cancel - self-interaction correction (SIC)
Self-interaction correctionHartree-Fock EJ + EX
F30 (orbital)
12 ij ij!"i, j# $ ij ji %&
i = j
Kohn-Sham EJ + EXC (density)12 !(x1) 1r12"" !(x2 )dx1dx2 + dx!(x)VXC !,#![ ]"
12 ii ii + i VXC !,"![ ] i # 0
Coulomb and exchange do not cancel - self-interaction error
* i*(x1) j!(x2 )"" 1
r12i (x1) j (x2 )dx1dx2 = ij ij
ii ii!" # ii ii $% = 0
© Joachim Sauer, HU Berlin, 2011
kinet. e-n e-e e-e en. attract Coulomb exchange/corr orbital density density
EHF = ET + EN + EJ + EXFock30 (orbital)
EDFT = ET + EN + EJ + EXC (density)functional of density onlyEXC
LDA = EXD30 + EC
VWN
functional also of density gradient (Generalized Gradient Approximation)EXC
BLYP = EXDirac30 + ΔEX
B88 + ECLYP
orbital-density hybrid functionalEXC
B3LYP = a0 EXF30 + (1- a0) EX
D30 + aX ΔEXB88
+(1- aC ) ECVWN + aC EC
LYP
Energy as functional of orbitals or electron density
Self-interactioncancels
Self-interactioncancels partially
©Joachim Sauer, HU Berlin, 2011
SOMO
20% 0% 50% Fock-exchange in functional
Cluster anions (V2O5)n-
Size-dependentelectron localization
©Joachim Sauer, HU Berlin, 2011
V6O15-
Cs- 2Aʻ-10
C2v- 2A2- 9
- 48 kJ/mol
BH-LYPCCSD(T)//BH-LYPB3-LYP
- 57 kJ/mol
Cs- 2Aʻʻ
D2d- 2B1
Cs- 2Aʻʻ
V4O10-
+ 33
CCSD(T) calculations confirm most stable structures
© Joachim Sauer, HU Berlin, 2011
Orbital functional for electron correlation:2nd order Møller-Plesset perturbation theory
ECMP2 ![ ] = "
ab rs 2
#r + # s " #a " #br<s$
a<b$
Covers dispersionFunctionals including this term and Fock exchange: double hybrid
EDFT = ET + EN + EJ + EXC [ρ] + ax EXF30 + aMP2 EC
MP2
© Joachim Sauer, HU Berlin, 2011
! = !o + Cia!i
a
i,a" + Cij
ab!ijab
i< j ,a<b
"
+ Cijkabc!ijk
abc +…i< j<k ,a<b<c
"
Wave function expansion
© Joachim Sauer, HU Berlin, 2011
Wave function expansion
! = Cn!nn" = !o + Ci
a!ia
i,a" + Cij
ab!ijab
i< j ,a<b
" + Cijkabc!ijk
abc +…i< j<k ,a<b<c
"
Result depends on- configuration space „full configuration interaction“- size of one-particle space (M, given by the basis set)
how many configurations of a given substitution level ?
M !N !(M ! N )!
= 4 "1011 for CH4 DZP basis M=70, N=10
© Joachim Sauer, HU Berlin, 2011
M-N (virtual space)
pVDZ
© Joachim Sauer, HU Berlin, 2011
HF/small basis set
HF/large basis set
HF-limit
El.corr.calculated
Accurate, non-rel.
Exact (=exp.)
El. correlationsenergy (Def.)
El. corr. energyin quantum chemical
practise
Energy
© Joachim Sauer, HU Berlin, 2011
Electron correlation - „Post Hartree Fock“
! = !o + Cia!i
a
i,a" + Cij
ab!ijab
i< j ,a<b
" + Cijkabc!ijk
abc +…i< j<k ,a<b<c
"
Ecorr = E ! EHF = Cabrs "0
r<sa<b
# H "abrs = Cab
rs
r<sa<b
# ab rs
! a*(x1)! b
"(x2 )# 1r12! r (x1)! s (x2 )dx1dx2 = ab rsab rs = ab rs ! ab sr
You only need double substitution to calculate the exact correlation energy,but to determine the doubles coefficients exactly, you need all substitutions
© Joachim Sauer, HU Berlin, 2011
Many body perturbation theory (MBPT)
H0 = F(! )!"
#0 $ - Determinant consisting of canonical HF orbitals
E(0) = %ii" E (0) + E (1) = EHF
Ecorr = E(2) = !
ab rs"r + " s ! "a ! "br<s
#a<b# ab rs = EMP2
!0 H !ar = 0 (BrillouinTheorem)
!0 V !abrs = !0 H !ab
rs = ab rs = cµac"bc# rc$sµ"#$% µ" #$
EXCMP2 = EX
F30 + ECMP2
© Joachim Sauer, HU Berlin, 2011
Limiting value
Typical oscillating behaviour of results obtained with the MP method
Energy (Property)
SCF
MP3
MP4MP2
© Joachim Sauer, HU Berlin, 2011
Coupled Cluster Expansion
! = eT"0 = (1+ T + 12 T2 + 1
3! T3 + ....)"0
eT =1+ T1 + (T2 + 12 T1
2) + (T3 + T2T1 + 13! T1
3) + (T4 + 12 T2
2 +
T1!0 =i" ti
m
m" !i
m
T12!0 =
i,j" ti
mtjn
m,n" ! i j
mn
T2!0 = t i jmn
m,n" ! i j
mn
i,j"
disconnected (unlinked) doubles
connected (linked) doubles
t - Amplitudes
The presence of disconnected (double) substitutionensures size-consistency
© Joachim Sauer, HU Berlin, 2011
Coupled Cluster Energy
Ecorr =i < j! (t ijmn + timt jn "
m<n! tint jm ) ij mn
Equations for determination of „amplitudes“ (coefficients)
(if HF orbitals are used)
timim! "l
p H"im + (t ijmn + timtjn # tintjm)
ij,mn! "l
p H" ijmn +
(t ijmntko + ...timtjntko # ...)! "lp H" ijk
mno = Etlp
Iterative solution
E= !0 H 1+T1+(T2 + 12 T1
2 )!0
© Joachim Sauer, HU Berlin, 2011
CC-SD equation (HF-Orbitale)
!im H (T1 + T2 + 1
2 T12 + T1T2 + 1
6 T13 )!0 = E " ti
m
! ijmn H (1+ T1 + T2 + 1
2 T12 + T1T2 + 1
6 T13 + 1
2 T22 + 1
2 T12T2 + 1
24 T14 )!0
= E " (t ijmn + timtjn # tintjm)
CCSD(T) - (T)riple-contribution according to MP4 equation,but T1 and T2 amplitudes are taken from CCSD equations
© Joachim Sauer, HU Berlin, 2011
Wavefunction-based methods for electron correlation
I. According to the way, the expansion coefficients are determinedvariational: Configuration Interaction (CI methods)
Size-consistency problemCIS (CI with singles) for excited states
perturbation theory: Møller-PlessetMP2, MP4-SDTQ
system of equations: Coupled Cluster (iterative)CCSD; CCSD(T)
© Joachim Sauer, HU Berlin, 2011
II. Substitutions included in wavefunction expansionall in given one-particle basis: full CIup to quadruple: MP4-SDTQup to connected triple substitution: CCSD(T)
(an expensive standard method)up to doubles: CI-SD → CPFdoubles only: MP2
(the simplest wavefunction-based standard method))singles only: CI-S
(simplest method for excited states)
Wavefunction-based methods for electron correlation
© Joachim Sauer, HU Berlin, 2011
Neese et al., Acc. Chem. Res. 42 (2009) 641-648
© Joachim Sauer, HU Berlin, 2011
The scaling problem
© Joachim Sauer, HU Berlin, 2011
Pople school
(GAUSSIAN)
correlation consistent
Dunning
6-311G**
6-311+G(2df,p)
6-311+G(3df,2p)
cc-pVDZ
cc-pVTZ
cc-pVQZ
3s2p1d/2s1p
4s3p2d1f/3s2p1d
5s4p3d2f1g/4s3p2d1f
+ diffuse s-function
aug - diffuse functions (one for each
angular momentum)
Polarisation functions indispensable/diffuse functions recommended
Electron correlation: requires better (larger) basis setsthan Hartree Fock or DFT
© Joachim Sauer, HU Berlin, 2011
Dunning (aug)-cc-pVnZ Basis Sets
s p d f g h No.AO X/H
cc-pVDZ 3/2 2/1 1/0 14/05cc-pVTZ 4/3 3/2 2/1 1/0 30/14cc-pVQZ 5/4 4/3 3/2 2/1 1/0 55/30 cc-pV5Z 6/5 5/4 4/3 3/2 2/1 1/0 91/55
aug-cc-pVDZ 4/3 3/2 2/0 23/09aug-cc-pVTZ 5/4 4/3 3/2 2/0 46/23aug-cc-pVQZ 6/5 5/4 4/3 3/2 2/0 80/46aug-cc-pV5Z 7/6 6/5 5/4 4/3 3/2 2/0 127/80
l/k refers to first row atom/h atom
© Joachim Sauer, HU Berlin, 2011
How to reach the basis set limitfor a given method?
• Use basis sets of increasing size• Reaching the basis set limit faces the scaling problem• More expensive methods generally need larger basis sets
to reach limit• Extrapolation formulas available
© Joachim Sauer, HU Berlin, 2011
Basis set extrapolation - CBS limit
(aug)-cc-pVXZ
Two-point extrapolationA. Halkier et al.,CPL 286 (1988) 243
other and more sophisticated schemes, see refs. inT. H. Dunning, Jr., JPC A 104 (2000) 9062
Note: You need to extrapolate also the Hartree-Fock energy
© Joachim Sauer, HU Berlin, 2011
O2 dissociation energy (kJ/mol)
Method Basis set De D0 Obsd 505 494 UCCSD(T)FC cc-pVTZ 476
cc-pVQZ 490 CBS(T,Q) 501 aug-cc-pVTZ 481 CBS 501 CBS+cv 503
G2 G2 494 484
EMSL Computational Results DataBase - 255 Molecules, 41 Atomshttp://www.emsl.pnl.gov:2080/proj/crdb/
© Joachim Sauer, HU Berlin, 2011
Au-S bond distance (pm) and H2S-Au interaction energies
(kJ/mol)
RAu-S E
PBE/TZVPP 247 -53.8
PBE+Db/TZVPP//PBE 247 -56.9
B3LYP/TZVPP//B3LYP 260 -26.5
B3LYP+Db/TZVPP 260 -30.7
CCSD(T)//B3LYP+Da E E/CBS(X,Y)
aug-pVDZ -27.4 -
aug-pVTZ -32.8 -34.7 (D,T)
aug-pVQZ -35.4 -37.0 (T,Q)
aug-pV5Z -36.7 -37.7 (Q,5)
b Au C6 parameters from Tonigold and Gross ON TOPON TOPFLATFLATAuAu
Au_adAu_adSS
GAS PHASEGAS PHASE
H HS
Au
© Joachim Sauer, HU Berlin, 2011
Basis Set Superposition Error
ΔE = EAB - ( EA + EB )
ΔEc = EAB - ( EA(B) + E(A)B )
εA = (EA - EA(B)) εB = (EB - E(A)B)
Boys Bernardi function counterpoise method
© Joachim Sauer, HU Berlin, 2011
BSSE
BSSEBasis set superposition errorBSCEBasis set convergence error
Thom H. Dunning Jr."A road map for the calculationof molecular binding energies"J Phys Chem A 2000, 104, 9062-9080
© Joachim Sauer, HU Berlin, 2011
Notes on BSSE
• BSSE for two subsystems is an indicator of balanced basis sets• For small basis sets BSSE correction may lead to larger
deviations from experiment (or basis set limit)• BSSE corrected results are less basis set dependent• There may be other errors (BSCE)• Sizeable for intermolecular complexes, but also present for
chemical bonds• BSSE elimination is not always easy. How do you correct an
energy barrier for BSSE?• There is no BSSE for plane wave basis sets
© Joachim Sauer, HU Berlin, 2011
There are cases, for which the Hartree Fock determinant is not asufficiently good starting pointStretched, almost dissociated bonds (H2 example)Biradicals, polynuclear transition metal compoundsAlmost degenerate states (partially occupied d-shells)
Multireference casesCAS (Complete Actice Space) SCF
MC (Multi-configuration) SCF+ MR (Multireference)-CI(SD)
©Joachim Sauer, HU Berlin, 2011
GGA functionals (PBE) are often not good enough, buthybrid functionals (B3LYP, PBE0, HSE) are one order of magnitude moreexpensive (with plane waves and pbc two orders of magnitude)
GGA functionals yield more delocalized electronic structures(delocalization of holes), preference for polar structures and too lowenergy barriers.
Partially occupied d-shells, the presence of more than one TM site andthe decoupling of electron pairs in bond breaking processes requires amulti-reference treatment- DFT limited to high-spin, low spin states require broken symmetryapproximation- in general, CCSD(T) is not good enough and not suitable as benchmark- ab initio multi-reference methods cannot treat large enough systemsreliably.
The challenge of transition metal oxides
© Joachim Sauer, HU Berlin, 2011
T1 diagnostics
Is CCSD(T) applicable?Is a single determinant a good reference function?
T1 > 0.02 not a clear single reference state
T1!0 =i" ti
m
m" !i
m
T1 = t1 / Ne t1 =i! (ti
m
m! )2
!= 1[ +T1+(T2 + 12T1
2 )+ (T2T1 + 13!T1
3 )+ ...]"0
©Joachim Sauer, HU Berlin, 2011
V2O5/SiO2 OV(OCH3) OV(OCH3)+
Calculated broken symmetry doublet
E#,B3LYP 154 147 80E#,CCSD(T) (137) 131ObservedE#
intrinsic (0 K) 137±10 52±11estimate
VV(d0)VIV(d1) - biradicaloid
broken symmetry VIII(d2)- triplet
H-Transfer: surface complex - gas phase model
Döbler, Pritzsche, Sauer, JACS 127 (2005) 10861
T1=0.07
© Joachim Sauer, HU Berlin, 2011
If you would like to join our group aspostdoc or PhD student -
(www.chemie.hu-berlin.de/ag_sauer